Fix register usage conventions for amd64 unfilter

modified:   stub/src/amd64-win64.pep.S
	modified:   stub/src/arch/amd64/bxx.S

src/stub/src/amd64-win64.pep.S

@@ -234,11 +234,10 @@ section         PEFILTER49
                 mov     ecx, offset filter_length
                 mov     dl, IMM8(filter_cto)
 .att_syntax
-#define esi ebx
+        push %rsi  // save
-#define rsi rbx
+        push %rbx; pop %rsi  // input len
 #include "arch/amd64/bxx.S"
-#undef esi
+        pop  %rsi  // restore
-#undef rsi
 .intel_syntax noprefix
 
 // =============

src/stub/src/arch/amd64/bxx.S

@@ -35,11 +35,9 @@ amdbxx:  # (*f_unf)(xo->buf, out_len, h.b_cto8, h.b_ftid);
 
 #ifndef NO_METHOD_CHECK
         cmpl $0x49,ftid; jne ckend0  # filter: JMP, CALL, 6-byte Jxx
-#endif
         push %rbx  # save
-	cld  # convention should have guaranteed this already
+#endif
-
+        push %rdi; lea -4(%rdi,%rsi),%rcx  # last possible displacement
-        push %rdi; lea (1- 4)(%rdi,%rsi),%rcx  # beyond last possible opcode
         pop  %rsi  # start of buffer
         push %rsi
         pop  %rbx  # remember start of buffer
@@ -56,7 +54,7 @@ ckloop2:
         subb $     0xE8,%al
         cmpb $0xE9-0xE8,%al; ja ckloop4  # not JMP, not CALL
 ckmark:
-        cmpq %rcx,%rsi; jae ckend  # peek only; not marked ==> do not consume
+        cmpq %rcx,%rsi; ja ckend  # peek only; not marked ==> do not consume
         push %rsi; lodsl  # (assume) marked, bswapped 32-bit displacement
         subb %dl,%al; jne ckloop3  # not marked with cto8
         pop %rdi
@@ -68,9 +66,9 @@ ckstart:
         cmpq %rcx,%rsi; jae ckend
         lodsb; jmp ckloop2  # 0x0F prefix would overlap previous displacement
 ckend:
+#ifndef NO_METHOD_CHECK
         pop %rbx  # restore
 ckend0:
-#ifndef NO_METHOD_CHECK
         ret
 #endif